Scholarlink

<p>We introduce ECT-Bench, the first large-scale, experimentally acquired dataset for planar Electrical Capacitance Tomography (ECT) designed to evaluate both imaging and dynamic tracking algorithms. The dataset contains 54,739 high-resolution mutual capacitance matrices acquired via a custom 3&#xa0;×&#xa0;3 electrode array. Taking advantage of a unified robotic pipeline, data was collected under comprehensive spatial variables, covering contact and multi-distance non-contact conditions (<i>Z</i>-axis permutations), alongside continuous spatio-temporal trajectories. Three representative materials (glass, resin, wood) and geometric shapes (circle, triangle, square) are thoroughly documented. Each sample couples raw capacitance signals with strict positional metadata and ground-truth object labels. Furthermore, ECT-Bench provides quantitative reconstructions comparing traditional physical solvers (Linear Back Projection, Landweber) against data-driven deep learning (CNNs). Unlike existing simulation-based benchmarks, ECT-Bench captures genuine soft-field characteristics, dielectric variability, positional uncertainty, and hardware noise. By openly releasing all sensor fabrication files, acquisition codes, and baselines, ECT-Bench establishes a standardized, rigorous foundation for bridging the gap between numerics and practical real-world touchless sensing.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Electrical Capacitance Tomography Dataset for Image Reconstruction Benchmarking

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月19日

查看全文
  • Duanpeng Shi,
  • Yuliang Wang,
  • Xu Li,
  • Tengchen Sun,
  • Huaping Liu,
  • Di Guo

摘要

We introduce ECT-Bench, the first large-scale, experimentally acquired dataset for planar Electrical Capacitance Tomography (ECT) designed to evaluate both imaging and dynamic tracking algorithms. The dataset contains 54,739 high-resolution mutual capacitance matrices acquired via a custom 3 × 3 electrode array. Taking advantage of a unified robotic pipeline, data was collected under comprehensive spatial variables, covering contact and multi-distance non-contact conditions (Z-axis permutations), alongside continuous spatio-temporal trajectories. Three representative materials (glass, resin, wood) and geometric shapes (circle, triangle, square) are thoroughly documented. Each sample couples raw capacitance signals with strict positional metadata and ground-truth object labels. Furthermore, ECT-Bench provides quantitative reconstructions comparing traditional physical solvers (Linear Back Projection, Landweber) against data-driven deep learning (CNNs). Unlike existing simulation-based benchmarks, ECT-Bench captures genuine soft-field characteristics, dielectric variability, positional uncertainty, and hardware noise. By openly releasing all sensor fabrication files, acquisition codes, and baselines, ECT-Bench establishes a standardized, rigorous foundation for bridging the gap between numerics and practical real-world touchless sensing.